www.nature.com/scientificreports OPEN Lesions in the right Rolandic operculum are associated with self‑rating afective and apathetic depressive symptoms for post‑stroke patients Stephanie Sutoko1,2*, Hirokazu Atsumori1,2, Akiko Obata1,2, Tsukasa Funane1,2, Akihiko Kandori1,2, Koji Shimonaga3,4, Seiji Hama2,4, Shigeto Yamawaki5 & Toshio Tsuji6 Stroke survivors majorly sufered from post‑stroke depression (PSD). The PSD diagnosis is commonly performed based on the clinical cut‑of for psychometric inventories. However, we hypothesized that PSD involves spectrum symptoms (e.g., apathy, depression, anxiety, and stress domains) and severity levels. Therefore, instead of using the clinical cut‑of, we suggested a data‑driven analysis to interpret patient spectrum conditions. The patients’ psychological conditions were categorized in an unsupervised manner using the k‑means clustering method, and the relationships between psychological conditions and quantitative lesion degrees were evaluated. This study involved one hundred sixty‑fve patient data; all patients were able to understand and perform self‑rating psychological conditions (i.e., no aphasia). Four severity levels—low, low‑to‑moderate, moderate‑ to‑high, and high—were observed for each combination of two psychological domains. Patients with worse conditions showed the signifcantly greater lesion degree at the right Rolandic operculum (part of Brodmann area 43). The dissimilarities between stress and other domains were also suggested. Patients with high stress were specifcally associated with lesions in the left thalamus. Impaired emotion processing and stress‑afected functions have been frequently related to those lesion regions. Those lesions were also robust and localized, suggesting the possibility of an objective for predicting psychological conditions from brain lesions. Structural brain abnormalities caused by brain infarction and hemorrhage bring complex impairments related to physical-cognitive functions and psychological conditions. Post-stroke depression (PSD) is closely linked to (afective) depression and apathy symptoms. Patients lacking self-acceptance due to stroke-evoked disabilities and having irrational expectations of their recovery course (high insistence on recovery) may develop both symp- toms. Even though both result in negative efects on a patient’s quality of life 1,2, those symptoms are considered distinct and separate domains. A previous study based on computed tomography imaging reported that lesion location may afect diferent symptoms. Te severity of depression was related to lesions in the lef frontal lobe; the symptomatic apathy was associated with the damage of bilateral basal ganglia3. Tose symptoms afected serotonergic and dopaminergic neurotransmitter pathways diferently 4, and efects of those symptoms on lesions were also less likely overlapped (12–21%; brainstem lesions). Te improvement of physical abilities was claimed to reduce depressive and apathetic symptoms2. However, the apathy domain was a better predictor of functional recovery than the depressive symptoms5. More than 50% of reported stroke patients sufered from apathy and/or depression, with varied possibilities of apathy without depression (20–28%), depression without apathy (12–20%), and both psychological symptoms 1Center for Exploratory Research, Research and Development Group, Hitachi. Ltd., Tokyo, Japan. 2Department of Rehabilitation, Hibino Hospital, Hiroshima, Japan. 3Department of Neurosurgery and Interventional Neuroradiology, Hiroshima City Asa Citizens Hospital, Hiroshima, Japan. 4Department of Neurosurgery, Graduate School of Biomedical and Health Sciences, Hiroshima University, Hiroshima, Japan. 5Center for Brain, Mind and KANSEI Sciences Research, Hiroshima University, Hiroshima, Japan. 6Graduate School of Advanced Science and Engineering, Hiroshima University, Hiroshima, Japan. *email: [email protected] Scientifc Reports | (2020) 10:20264 | https://doi.org/10.1038/s41598-020-77136-5 1 Vol.:(0123456789) www.nature.com/scientificreports/ Figure 1. Plot of explained variance against cluster number in categorizing patient data based on their apathy and depression scores. By using the elbow method, the optimum cluster number was found to be four. (15–21%) across sites 2–4. Te variations of psychological condition and even afected lef hemispheric lesion might be caused by methodological diferences across studies 6. Te diagnostic discrepancy is likely to have been a result of using divergent types of psychometric inventories with particular cut-ofs. Furthermore, the applied cut-of interpreted psychological condition from a binary perspective of severity level (diagnosed vs. non-diagnosed; low vs. high). Here, however, we argue for a higher-order interpretation on the basis of multiple severity levels (i.e., a spectrum) of apathetic-depressive symptoms or even other psychological domains (e.g., anxiety, stress). As discussed above, the implementation of cut-ofs on patient’s scores of psychometric inventories may bring about a limited interpretation for the dataset. Terefore, rather than using the pre-set clinical cut-ofs, we suggest data-driven and unsupervised categorization of patient data based on two psychological domains. Te existent of lesions had been associated with psychological conditions as mentioned above, but the relationship between a spectrum of severity levels and quantitative lesion degrees has not been studied before. Terefore, in the current study, we aimed to understand the spectrum of psychological domains and severity levels for stroke patients and its relationships with brain lesion degrees. Results Four clusters interpreting psychological conditions for stroke patients. Unsupervised categori- zation requires the pre-determined cluster number. In order to avoid any assumptions, the cluster number was optimized. Te optimization target was the clustering efciency evaluated by a parameter, namely, explained variance (range 0–100%). As the cluster number increases, the explained variance parameter gradually increases and reaches a plateau as shown in Fig. 1. Te diference of explained variance for two sequential cluster numbers was initially high (for the increase from two clusters to three) and gradually decreased (for the increases from three clusters to four, four to fve, and so on). Te smallest cluster number (i.e., two clusters) revealed around 87% explained variance; the largest cluster number (i.e., ten clusters) brought almost 100% explained variance. When the increase of explained variance becomes insignifcant, further increasing the cluster number will not result in better clustering. In order to determine the optimum cluster number, the elbow method7,8 was applied on the plot of explained variance against cluster number. A frst-degree line between the most-distant points (two and ten clusters) was ftted. Te distance between points ([cluster number, explained variance]; red dots in Fig. 1) and the line was computed. Te longest distance was obtained by the point of four clusters (98% explained variance). Terefore, four was selected as the optimum cluster number. Te optimum cluster number for clas- sifying patient data based on other combinations of psychological domains (e.g., apathy–anxiety, depression– anxiety, and so on) was also four clusters (data not shown). Distinct lesion characteristics for each cluster. Te patient data on depression-plotted-against-apathy scores were categorized into four clusters. Forty-two, 51, 50, and 22 patient data were categorized for clusters 1, 2, 3, and 4, respectively. Figure 2 shows the lesion maps visualized in six ways—anterior, posterior, top, bottom, lef, and right views. Te colored patches on brain identify the lesion degrees averaged across patients. Tese maps improved the process of exploratory data analysis. Te lesion characteristics were distinct for each cluster. While all clusters revealed major lesions in the occipital lobe (Fig. 2A2–D2), the prefrontal lesions were observed only for clusters 2 and 3 (Fig. 2B1, C1). Cluster 2 was distinguished from cluster 3 based on lesions found in the lef precentral–postcentral–superior parietal gyri (Fig. 2B3 vs. C3). Clusters 1 and 4 revealed lesions in the orbital part of inferior frontal gyrus with specifc lef (Fig. 2A4) and right (Fig. 2D4) laterality for respective clusters. Besides the lesion locations, the lesion degree (percentage) may be useful to explain cluster characteristics. For example, even though both clusters 1 and 4 showed lesions in the middle and inferior temporal gyri, a greater lesion degree was observed for cluster 4 (Fig. 2C6 vs. D6). Scientifc Reports | (2020) 10:20264 | https://doi.org/10.1038/s41598-020-77136-5 2 Vol:.(1234567890) www.nature.com/scientificreports/ Figure 2. Lesion maps for each clusters (A–D for clusters 1, 2, 3, and 4, respectively) visualized in anterior (A1– D1), posterior (A2–D2), top (A3–D3), bottom (A4–D4), lef (A5–D5), and right (A6–D6) views. Color bar represents the lesion degree in percentage. A, P, R, and L denote anterior, posterior, right, and lef, respectively. Cluster efect on brain lesion degree at the right Rolandic operculum and the left thala‑ mus. Te psychological domains are displayed in 2-axis scatter plots in which each axis represents a psy- chological domain (Fig. 3). Tere are two clustering characteristics. First, each psychological domain was cat- egorized into four severity levels, such as low, low-to-moderate, moderate-to-high, and high. Clusters then Scientifc Reports